Nonlinear knowledge in learning models
Roberto Prevete · 2007
For most real life problems it is difficult to find a classifier with optimal accuracy. This motivates the rush towards new classifiers that can take advantage of the experience of an expert. In this paper we propose a method to include nonlinear prior knowledge in Generalized Eigenvalues Support Vector Machine. The expression of nonlinear kernels and nonlinear knowledge as a set of linear constraints allows us to have a nonlinear classifier which has a lower complexity and halves the misclassification error with respect to the original generalized eigenvalues method. The Wisconsin Prognostic Breast Cancer data set is used as a case study to analyze the performance of our approach, comparing our results with state of the art SVM classifiers. Sensitivity and specificity results for some publicly available data sets well compare with the other considered methods.